基于XGBoost算法的摇床分选指标预测研究  

Prediction of Separation Indices of Shaking Table Based on XGBoost

作　　者：何微 戴昊霖 陈雯[1] HE Wei;DAI Haolin;CHEN Wen(Changsha Research Institute of Mining and Metallurgy Co.,Ltd.,Changsha 410012,Hunan,China;School of Automation,Central South University,Changsha 410083,Hunan,China)

机构地区：[1]长沙矿冶研究院有限责任公司,湖南长沙410012 [2]中南大学自动化学院,湖南长沙410083

出　　处：《矿冶工程》2025年第1期70-75,共6页Mining and Metallurgical Engineering

摘　　要：提出了一种基于极限梯度提升(XGBoost)算法的摇床分选指标预测方法。该方法通过颜色矩量化矿带的颜色特征,图像矩量化矿带的形态特征,灰度共生矩阵评价指标的对比度、同质性、相关性和ASM能量量化矿带的纹理特征,根据分选过程中矿带颜色、形态和纹理的变化有效提取摇床矿带特征;然后将图像特征作为输入、选矿指标作为输出,利用XGBoost算法筛选矿带特征并构建预测模型,训练模型并在测试集上预测精矿品位、回收率和精矿产率,实现对分选指标的准确预测。通过与决策树模型和随机森林模型的精矿品位、精矿产率和回收率预测结果对比,基于XGBoost算法的模型对精矿回收率、产率预测精度高。A prediction method for mineral processing indices of shaking table was proposed based on eXtreme Gradient Boosting(XGBoost).As for the separation product zone,it can quantify the colour difference by colour moments,the shape characteristics by image moments,and the texture characteristics by evaluation indices of grey-level co-occurrence matrix,including contrast,homogeneity,correlation and ASM energy.The features of the separation product zone can be effectively extracted according to the difference in its colour,shape and texture during separation process.Subsequently,those features can be filtered by employing XGBoost,with which a prediction model can be then constructed.After being trained with the test set to predict the grade,recovery rate and yield of concentrate,this model can achieve accurate prediction of separation indices.It is found that the XGBoost-based model outperforms the decision tree model and random forest model in terms of accuracy when applied for predicting recovery and yield of concentrate.

关 键 词：摇床 特征提取 品位预测 XGBoost 特征选择 图像识别 智能控制 分选指标 预测模型 矿带 

分 类 号：TD922[矿业工程—选矿] TP274[自动化与计算机技术—检测技术与自动化装置]